Imaging genomics has emerged recently as a very promising and active research area by combining imaging and genomics approaches for comprehensive and systematic diagnosis of complex diseases. Utilizing multiscale and multimodal imaging genetic techniques such as fMRI imaging and SNP arrays, complementary information can be fused for better diagnosis and prognosis of diseases. However, fusion of these heterogeneous data has been extremely difficult. Most current approaches still analyze these data separately or with simple pairwise correlations and regression; many significant challenges exist: 1) Many available biological knowledge databases (e.g., protein-protein interactions (PPI)) contain rich and useful information but they have not been incorporated into data fusion. 2) Currently available data fusion approaches usually overlook the inter-correlations between imaging and genetics data, especially the interaction patterns within/between each type of data. 3) Imaging genetic data usually have smaller sample size but contain greater number of features. Many current approaches fail to consider these specific features and become ineffective in processing these data. To this end, the goal of this project is to tackle above significant challenges by developing innovative data integration approaches for the detection of novel biomarkers and use them for the identification of genes (modules) and improved diagnosis of complex diseases. We have assembled a multidisciplinary team including bioinformatician and biomedical engineer, imaging scientist, statistical geneticist, clinical psychiatrist, and medical informatician with complementary and synergistic expertise and experiences. We have collaborated productively and our preliminary results have demonstrated promising results for improved diagnosis of disease with integrative approaches. Building upon this success, we plan to accomplish the following specific aims: to develop novel computational approaches to correlate and integrate fMRI imaging with genomic data while incorporate biological knowledge and their interaction networks for the detection of biomarkers; and to apply/validate the detected biomarkers for the identification of risk genes/gene modules and for the improved diagnosis of subtle patient subgroups. Through this project, we will deliver a set of powerful sparse model based methods for imaging and genomic data fusion, especially by incorporating interaction networks and biological knowledge, which are often overlooked by current approaches. In addition, we will disseminate the developed methods via an open source software toolbox so that this project can have a broad and sustainable impact. We use mental disorders as a prototype for the validation but the developed models and tools can be applicable to studies of multiple other diseases.